To our knowledge, this is one of the first studies analyzing the influence of SEP on the relationship between physical health conditions and SRH on the one hand and a generic measure of HRQoL, the SF-36 questionnaire, on the other hand. Our results suggest that SEP influences the impact of health conditions, like FL or chronic low back pain, on subjective health in a different way according to whether it is measured by SRH or HRQoL. Compared with people with low SEP, some health conditions like chronic low back pain seem to have a greater negative impact on SRH in socially advantaged people, but the opposite occurred for quality of life. The strength of this interaction varied according to the indicator used to measure health conditions as well as the indicator used to define SEP.
An important limitation of our study is that health conditions were self-reported, and may be susceptible to misreporting . However, we mainly studied chronic disabling diseases, assessed using valid questionnaires or a standardized definition that may be less susceptible to this type of misreporting . Moreover, self reports could be reasonably accurate for certain chronic conditions [39, 40]. Haapanen et al. showed that agreement between questionnaire data and medical records may be good for chronic diseases that have a clear definition . As we used chronic health conditions and valid questionnaires or definitions to measure them, we believe that the proportion of misreporting is low, and unlikely to explain the opposite directions for SRH and HRQoL. Idler et al. showed that knowledge of a chronic illness strengthened the association between SRH and mortality . Thus, use of patient-reported health conditions could constitute an appropriate indicator for analyzing the association between health conditions and subjective health. However, future studies are needed to examine the influence of SEP on the relationship between subjective health and objective health or "true health", assessed through more objective measures or by using multiple indicators linear structural equation models with latent variables as done by Shmueli et al. .
Another limitation is that some items used to evaluate FL are components of PCS score of the SF36 questionnaire. As low educated people have higher number of FL than high educated men, their PCS score should be poorer and could explain why PCS scores are poorer among lower educated people in case of disease. Although FL is subsumed within the concept of HRQoL measured by the SF-36 questionnaire, these two measures are not exactly the same. As an illustration, some works have shown that FL was a predictive factor of HRQoL, justifiying that FL and HRQoL are two different concepts [22, 25, 26]. In our study correlations between PCS score and FL were -0.38 in men and -0.49 in women. Therefore we do not think that this correlation is likely to explain totally the lower decrease of PCS score observed among people with high SEP. Moreover the same interaction is observed with MCS score for which no correlations were found between FL and MCS. Finally, we observed a lower decrease of PCS score for men with higher SEP in case of chronic low back pain, which is a different measure than the SF-36 questionnaire.
Another limitation is that tests of interaction have usually classically low power [44, 45]. It is thus likely that some of interactions tests we performed lacked power to put in evidence a significant influence of SEP on the relationship between health status and subjective health.
In our study we used the SF-36 scale as a measure of HRQoL. As observed by Luo et al. health status assessed by different HRQoL indicators is not exactly the same . Even more disturbing for the analysis of social inequalities in health, socioeconomic disparities may vary according to HRQoL indicators used to measure health and indicators used to measure socioeconomic position . Therefore, studies exploring HRQoL with indicators other than the SF-36 are needed.
This study's main strength lies in the fact that the NHS is a national and representative sample and enabled measurement of socioeconomic position by using education and income.
Socially advantaged people were generally at less risk of having or reporting health problems. As expected, SRH and HRQoL were positively associated with SEP. Regardless of health conditions people with lower educational attainment or lower income were more likely than others to report poor SRH and to have poorer quality of life. This gradient was consistent using any of the indicators of SEP, in contrast to the observations of Robert et al. who found that income was more consistently associated with HRQoL and SRH measures among US adults .
The presence of a given health condition lowered reported levels of SRH and HRQoL, but the relative impact this condition had on SRH and on HRQoL was different. Regarding SRH, the influence of SEP on the relationship between chronic health conditions and SRH was not consistently significant, but this influence was mainly in the same direction: the impact of chronic health condition, like chronic low back pain, was relatively greater for socially advantaged people. Put differently, people with a high SEP were more likely to report a negative impact of this health condition on their SRH than those with a low SEP. One possible explanation of this finding is that a person's expectations about their health increase with increasing SEP . The repercussions of health problems on SRH would therefore be worse for those with higher health expectations. Another possibility is that one's ability to be aware of one's own health status and to estimate risk is higher in socially advantaged people [48, 49]. In the event of disease, they are more likely to be aware of the consequences of a health problem, in terms of morbidity or mortality risks, and thus more likely to report poor self-rated health.
In contrast, regarding quality of life, the impact of health conditions on PCS and MCS was lower for socially advantaged people. Shmueli et al. had also showed that, for a same "true health state" (true health considered as a latent variable), individuals in better economic status reported higher health related quality of life than individuals in poor economic status . In our study, among men, the higher their income, the lower its impact on the PCS score, this interaction being less consistent with education. Among women, the same phenomenon was observed for education in case of chronic low back pain. It is noteworthy that income seems to have more influence on the relationship between health conditions and PCS score in men than in women. For MCS score, FL lessened the MCS score for the most highly educated and richest men. Among women, no influence of education was found but the impact on MCS score of chronic low back pain and FL was less pronounced in women with higher income. It is likely that the subjectivity is higher by using SRH, a single item on health in general, than with SF-36 questionnaire, which is a questionnaire with valid items focusing people on specific aspects of health. Therefore it may be a less subjective measure than SRH and less exposed to variability associated with individual health expectations. Moreover quality of life is a broader concept than SRH. Several dimensions of life are important, such as subjective well-being, happiness, life satisfaction or social relationships and networks . The notion of resources is probably in part at the origin of this contrast between the two indicators. While perceived health depends on expectations and on comparison with peers, quality of life refers back to an analysis close to that of handicap in opposition to incapacity. Quality of life estimates in a broad way how a disease or disability influences social functioning. In this respect, the notion of financial, social and cultural resources becomes essential to deal with the health conditions. In this context, a high level of resources could limit the impact of a disease on quality of life .